Statistical analyses
To analyse drivers of P. subordinaria epidemics in P. lanceolata populations in the Åland Islands, we ran Generalized linear models in SAS Proc Glimmix software (SAS Institute Inc.) with infection (0 = no infection, 1 = infection) as a binomial response variable, host population size, and connectivity as covariates, and regional district as categorical explanatory variable. Including host population connectivity allows controlling for spatial variation in putative gene flow among populations(Hanski, 1999), and including regional district allows controlling for possible regional variation in these data. To understand factors explaining P. subordinaria population size within-host populations, we fit a model defining the percentage of infected plants within-host populations as a response variable (1 ≤ 10% infected plants; 2 = 11-25% infected plants; 3 = 26-50% infected plants; 4 = 51-100% infected plants) with host population size and connectivity as covariates and regional district as a categorical explanatory variable. To understand how within-host infection load and disease transmission are linked in natural populations, we included pathogen within-host infection load (as average proportion of infected stalks in a population) as a covariate in the model. A Gamma distribution of errors was assumed.
We then analysed the results from the laboratory trial measuring the performance of two P. subordinaria strains alone and in coinfection with PlLV using generalized linear mixed models implemented in SAS Proc Glimmix software (SAS Institute Inc.) To test whether PlLV coinfection influences P. subordinaria lesion development (within-host infection load), we used AUDPS as a response variable with a Gaussian distribution of errors. In this model, we used P. subordinaria strain, virus inoculation treatment (1 = PlLV inoculation; 0 = mock inoculation), and time as categorical explanatory variables. Flower stalk (n = 92) was nested under plant individual (n= 32) and used as random factor. To understand how virus coinfection andP. subordinaria strain identity affect pycnidia formation, we analysed the subset of stalks from which the pycnidia were counted (n = 32) using a generalized linear model. Number of pycnidia was used as a continuous response variable and PlLV coinfection and P. subordinaria strain as categorical explanatory variables. Poisson distribution of error was assumed.
To test for possible trade-offs between within-host infection load (measured as AUDPS), and transmission potential, measured as pycnidia number, we ran a model using the stalks (n = 92) from which pycnidia were counted as generalized linear models in SAS Proc Glimmix software (SAS Institute Inc.). We included AUDPS as a covariate, and virus inoculation treatment (1 = PlLV inoculation; 0 = mock inoculation) and P. subordinaria strain as categorical explanatory variables. The response variable in the model was the pycnidia abundance as the pycnidia number counted in 1 cm2 area on the lesions. A Poisson distribution of errors was assumed.